This section details how to install OpenSearch on your host, including which operating systems are compatible with OpenSearch, which ports to open, and which important settings to configure on your host.
Operating system compatibility
OpenSearch and OpenSearch Dashboards are compatible with Red Hat Enterprise Linux (RHEL) and Debian-based Linux distributions that use
systemd, such as CentOS, Amazon Linux 2, and Ubuntu Long-Term Support (LTS). While OpenSearch and OpenSearch Dashboards should work on most Linux distributions, we only test a subset.
The following table lists the operating system versions that we currently support.
File system recommendations
Avoid using a network file system for node storage in a production workflow. Using a network file system for node storage can cause performance issues in your cluster due to factors such as network conditions (like latency or limited throughput) or read/write speeds. You should use solid-state drives (SSDs) installed on the host for node storage where possible.
The OpenSearch distribution for Linux ships with a compatible Adoptium JDK version of Java in the
jdk directory. To find the JDK version, run
./jdk/bin/java -version. For example, the OpenSearch 1.0.0 tarball ships with Java 15.0.1+9 (non-LTS), OpenSearch 1.3.0 ships with Java 220.127.116.11+1 (LTS), and OpenSearch 2.0.0 ships with Java 17.0.2+8 (LTS). OpenSearch is tested with all compatible Java versions.
|Compatible Java Versions
|Bundled Java Version
|8, 11, 14
|11, 17, 21
To use a different Java installation, set the
JAVA_HOME environment variable to the Java install location. For example:
The following ports need to be open for OpenSearch components.
|OpenSearch Dashboards in AWS OpenSearch Service with encryption in transit (TLS)
|OpenSearch REST API
|Node communication and transport
For production workloads, make sure the Linux setting
vm.max_map_count is set to at least 262144. Even if you use the Docker image, set this value on the host machine. To check the current value, run this command:
To increase the value, add the following line to
sudo sysctl -p to reload.
For Windows workloads, you can set the
vm.max_map_count running the following commands:
wsl -d docker-desktop
sysctl -w vm.max_map_count=262144
The sample docker-compose.yml file also contains several key settings:
Disables swapping (along with
memlock). Swapping can dramatically decrease performance and stability, so you should ensure it is disabled on production clusters.
bootstrap.memory_locksetting will cause the JVM to reserve any memory it needs. The Java SE Hotspot VM Garbage Collection Tuning Guide documents a default 1 gigabyte (GB) Class Metadata native memory reservation. Combined with Java heap, this may result in an error due to the lack of native memory on VMs with less memory than these requirements. To prevent errors, limit the reserved memory size using
-XX:MaxMetaspaceSizeand set the size of the Java heap to make sure you have enough system memory.
Sets the size of the Java heap (we recommend half of system RAM).
Sets a limit of 65536 open files for the OpenSearch user.
Allows you to access Performance Analyzer on port 9600.
Do not declare the same JVM options in multiple locations because it can result in unexpected behavior or a failure of the OpenSearch service to start. If you declare JVM options using an environment variable, such as
OPENSEARCH_JAVA_OPTS=-Xms3g -Xmx3g, then you should comment out any references to that JVM option in
config/jvm.options. Conversely, if you define JVM options in
config/jvm.options, then you should not define those JVM options using environment variables.
Important system properties
OpenSearch has a number of system properties, listed in the following table, that you can specify in
-D command line argument notation.
|By default, OpenSearch does not impose any limits on the maximum length of the JSON/YAML/CBOR/Smile string fields. To protect your cluster against potential distributed denial-of-service (DDoS) or memory issues, you can set the
opensearch.xcontent.string.length.max system property to a reasonable limit (the maximum is 2,147,483,647), for example,
|By default, OpenSearch serializes floating-point numbers using the default implementation provided by the Java Runtime Environment. Set this value to
true to use the Schubfach algorithm, which is faster but may lead to small differences in precision. Default is
|By default, OpenSearch does not impose any limits on the maximum length of the JSON/YAML/CBOR/Smile field names. To protect your cluster against potential DDoS or memory issues, you can set the
opensearch.xcontent.name.length.max system property to a reasonable limit (the maximum is 2,147,483,647), for example,
|By default, OpenSearch does not impose any limits on the maximum nesting depth for JSON/YAML/CBOR/Smile documents. To protect your cluster against potential DDoS or memory issues, you can set the
opensearch.xcontent.depth.max system property to a reasonable limit (the maximum is 2,147,483,647), for example,